Prev: How to surface plots over 3d models in MATLAB?
Next: Neural network train problem with validation
From: Andy Hennig on 7 May 2010 04:18
Hello,


I would like to calculate an autocorrelation for a quite long input-vector, at least longer than the magical 2^20 for which "xcorr" can be used.

In the documentary it says: "If you need to process longer sequences, see dfilt.fftfir." As this is not my field of expertise I looked for examples to use on the internet but couldn't find none.

Does anyone have any experience on how to calculate an autocorrelation on vectors longer than 2^20 numbers?

I'd appreciate any feedback on this,
eisvogel

xcorr: http://www.mathworks.com/access/helpdesk/help/toolbox/signal/xcorr.html
dfilt.fftfir: http://www.mathworks.com/access/helpdesk/help/toolbox/signal/dfilt.fftfir.html
From: Bruno Luong on 7 May 2010 08:39
Andy Hennig <phoenixasche(a)gmx.net> wrote in message <572026583.88016.1273234762059.JavaMail.root(a)gallium.mathforum.org>...
> Hello,
>
>
> I would like to calculate an autocorrelation for a quite long input-vector, at least longer than the magical 2^20 for which "xcorr" can be used.
>
> In the documentary it says: "If you need to process longer sequences, see dfilt.fftfir." As this is not my field of expertise I looked for examples to use on the internet but couldn't find none.
>
> Does anyone have any experience on how to calculate an autocorrelation on vectors longer than 2^20 numbers?

Well, I'l mot sure if the document is up-to-date. I try

>> x=rand(2^22,1); y=rand(size(x));
>> c=xcorr(x,y);
>> size(c) % 2^23-1 elements

ans =

8388607 1

It works just fine.

Bruno
From: Mark Shore on 7 May 2010 09:44
"Bruno Luong" <b.luong(a)fogale.findmycountry> wrote in message <hs11l8$5je$1(a)fred.mathworks.com>...
> Andy Hennig <phoenixasche(a)gmx.net> wrote in message <572026583.88016.1273234762059.JavaMail.root(a)gallium.mathforum.org>...
> > Hello,
> >
> >
> > I would like to calculate an autocorrelation for a quite long input-vector, at least longer than the magical 2^20 for which "xcorr" can be used.
> >
> > In the documentary it says: "If you need to process longer sequences, see dfilt.fftfir." As this is not my field of expertise I looked for examples to use on the internet but couldn't find none.
> >
> > Does anyone have any experience on how to calculate an autocorrelation on vectors longer than 2^20 numbers?
>
> Well, I'l mot sure if the document is up-to-date. I try
>
> >> x=rand(2^22,1); y=rand(size(x));
> >> c=xcorr(x,y);
> >> size(c) % 2^23-1 elements
>
> ans =
>
> 8388607 1
>
> It works just fine.
>
> Bruno

I was able to do the same with x=rand(2^25,1) using 2010a x64, although my computer hung with 2^26. Guess the documentation *is* out of date.
From: EE Student on 7 May 2010 12:40
Andy Hennig <phoenixasche(a)gmx.net> wrote in message <572026583.88016.1273234762059.JavaMail.root(a)gallium.mathforum.org>...
> Hello,
>
>
> I would like to calculate an autocorrelation for a quite long input-vector, at least longer than the magical 2^20 for which "xcorr" can be used.
>
> In the documentary it says: "If you need to process longer sequences, see dfilt.fftfir." As this is not my field of expertise I looked for examples to use on the internet but couldn't find none.
>
> Does anyone have any experience on how to calculate an autocorrelation on vectors longer than 2^20 numbers?
>
> I'd appreciate any feedback on this,
> eisvogel
>
> xcorr: http://www.mathworks.com/access/helpdesk/help/toolbox/signal/xcorr.html
> dfilt.fftfir: http://www.mathworks.com/access/helpdesk/help/toolbox/signal/dfilt.fftfir.html



Assuming x1 is input 1 and has length L1, x2 is input 2 and has length L2 are you speciffically looking for an output with with length L1+L2 - 1? If not you may be able to calulate a circular auto correlation such that cc = fftshift(real(ifft(fft(x1).*conj(fft(x1))))). I dont know if this will allow you to calculate vectors longer than2^25 but its worth a shot
From: Andy Hennig on 8 May 2010 04:29
Dear Nicholas,

you are right, I only need one side of the autocorrelation. Unfortunately, I am not familiar with the term "circular autocorrelation" nor with what your code does exactly, however I have tried what you suggested in comparison to "xcorr(x,y)" but it doesn't seem to be calculating an autocorrelation as I am used from "xcorr". I have uploaded a comparison: http://img265.imageshack.us/i/xcorrrlength1.png/

Or maybe I have applied your hint the wrong way?

Andy

P.S. Why is my email shown, even though in my settings I got it set to 'hidden'?
 |  Next  |  Last
Pages: 1 2
Prev: How to surface plots over 3d models in MATLAB?
Next: Neural network train problem with validation